From Experimental Signal to First-Class Continuous Profiling Database
Pyroscope 2.0 marks a significant evolution in continuous profiling, elevating it from a niche tool to a core component of modern observability. Continuous profiling complements metrics, logs, and traces by pinpointing exactly which function and line of code are consuming CPU cycles, enabling precise optimisations instead of blunt hardware upgrades. Grafana Labs rebuilt Pyroscope as an open-source continuous profiling database with a focus on making profiling at scale practical, not just possible. The release aligns with the broader industry trend of treating profiling as a standard signal: OpenTelemetry has added continuous profiling to its core telemetry model and declared its Profiles signal alpha. Pyroscope 2.0’s native support for the OpenTelemetry Protocol (OTLP) lets teams ingest profiles through familiar pipelines, lowering adoption friction and encouraging organisations to treat profiling as a default part of production monitoring rather than a rare, incident-only tool.
Storage Cost Reduction Through Single-Write Architecture and Symbol Deduplication
A central design goal of the Pyroscope 2.0 release is aggressive storage cost reduction. Earlier versions inherited a Cortex-based architecture that wrote each profile three times along the write path, multiplying storage use for payloads that can reach tens of megabytes. The new architecture eliminates write-path replication entirely: each profile is written once to object storage, which now acts as the single source of truth. Beyond write amplification, Pyroscope 2.0 targets symbolic data overhead. Profiles from the same service are co-located, allowing function names, source locations, and stack traces to be deduplicated across many profiles. In Grafana’s own production environment, this approach reportedly cut symbol storage footprints by up to 95%. For teams capturing continuous profiles across fleets of services, these optimisations translate directly into lower storage bills and a more predictable cost model, making always-on profiling feasible instead of a luxury reserved for critical hotspots.
Stateless Read Path Simplifies Operations and Handles Bursty Query Patterns
Pyroscope 2.0’s rearchitecture extends beyond storage, reshaping how queries are processed to reduce operational complexity. Previously, query execution lived inside stateful components that were hard to scale elastically. Capacity had to be provisioned for peak incident traffic, even though profiling queries often have no meaningful baseline and spike only during investigations. The new release makes the entire read path stateless: any querier can serve any request, and the number of queriers can scale up or down with demand. This elasticity helps absorb bursty query patterns, including automated access from LLM-powered agents that increasingly analyse profiling data during incidents. Fewer stateful components also mean fewer failure modes and faster, safer rollouts. Grafana reports that deployments that once took 8–12 hours now complete in minutes, a substantial reduction in operational friction for teams managing profiling at scale in complex environments.
Cleaner Data Model Unlocks New Profiling Workflows at Scale
The redesigned architecture is not just a cost-saving exercise; it also enables new ways of working with profiles. Pyroscope 2.0 introduces metrics derived from profiles, allowing teams to aggregate profiling data into fleet-wide comparisons without repeatedly querying individual profiles. This supports workflows such as tracking performance regressions across deployments or services. The system also now supports inspecting single profile instances instead of only aggregates, giving engineers fine-grained visibility when they need to drill into a specific incident. Heatmap queries provide visualisations of profile distributions over time, revealing patterns such as intermittent hotspots or gradual regressions. These capabilities emerge from a cleaner data model and stateless read path rather than bolt-on features, which means they can scale alongside the core system. Together, they make continuous profiling more actionable and better integrated into day-to-day performance engineering practices.
Positioning Pyroscope in a Growing Profiling Ecosystem
Pyroscope 2.0 arrives in a competitive and rapidly maturing ecosystem for profiling at scale. Open-source projects like Polar Signals’ Parca, which leverages eBPF and a custom time-series database (FrostDB), tackle similar storage and query efficiency challenges through different implementation strategies. On the commercial side, vendors such as Datadog, New Relic, Dynatrace, Sentry, and CubeAPM offer managed continuous profiling as part of broader observability platforms. Grafana’s positioning is distinctive: Pyroscope remains open source, while Grafana Cloud Profiles provides a hosted offering powered by the 2.0 architecture, which has already processed 19.5PB of profiling data in production. Real-world adopters like Monzo and Uber demonstrate how continuous profiling can detect regressions early and support profile-guided optimisation. For organisations already invested in the Grafana stack, Pyroscope 2.0 lowers both cost and operational barriers to adding continuous profiling to their observability strategy.